Model training

This section outlines the training stages of the pipeline, which includes training and evaluating both the Tn5 bias model and the bias-factorised ChromBPNet model.

These stages are centered around the predefined chrombpnet commands retrievable from the ChromBPNet repository. The commands used from this repository are chrombpnet bias pipeline and chrombpnet pipeline.

The reason for not directly using the pipeline directly from the original repository is solely due to the large scale of data to be processed from the Consensus Pituitary Atlas (Kövér et al., 2026). The following therefore outlines how these functions were adapted for this scale, using the adult mm10 cohort of scATAC-seq data as an example.